DataTerminology.comNEW
Crafting inputs to guide large language models toward desired behaviors or outputs.
DESCRIPTION
Prompting designs instructions, context, and examples to steer generative models. Effective prompts clarify task, constraints, and output format, often leveraging few-shot examples and system directives. Techniques include chain-of-thought, role assignment, and structured templates. Prompting complements fine‑tuning and RAG by improving performance without retraining. It is iterative: users test variants, evaluate responses, and refine. Good prompting reduces hallucinations and aligns outputs with user intent.
ALTERNATIVE DEFINITIONS
Wikipedia
Prompt engineering is the process of designing and refining prompts to obtain desired outputs from artificial intelligence (AI) systems such as large language models.
Emphasizes design and refinement for desired outputs across AI systems.
OpenAI
Prompt engineering is the practice of guiding models using constructed text inputs to achieve specific outcomes.
Frames prompting as practical guidance to achieve outcomes.
RELATED TERMS
Large Language Model Prompt Template Assistant Prompt Chain-of-Thought Few-Shot Learning Zero-shot Learning Temperature Tokenization Hallucination
COMPETING TERMS
Fine-Tuning Retrieval-Augmented Generation
RELATIONS
- Narrower: Chain-of-Thought Prompting Few-shot Prompting Role Prompting Zero-shot Prompting
- Referenced by: BIG-bench
KNOWLEDGE AREAS
AI Data Science MLOps Platform & Tooling
TARGET AUDIENCE
AI Engineer Data Scientist MLOps Engineer AI Product Manager Chief AI Officer Data Analyst
EXTERNAL SOURCES
- Wikipedia - Prompt engineering
- OpenAI - Best practices guide
- Anthropic - Prompting strategies
- Google Cloud - Prompt design guidance
TOP QUESTIONS
What is the goal of Prompting for LLMs?
Prompting aims to steer model behavior toward accurate, relevant, and well‑formatted outputs by providing clear instructions, context, and examples.
Category: Overview
How does Prompting compare to fine-tuning?
Prompting changes inputs, not model weights, yielding faster iteration and lower costs. Fine‑tuning adapts the model for persistent patterns but requires data and training.
Category: Alternatives
What are common mistakes in Prompting?
Vague instructions, missing constraints, and hidden assumptions often cause hallucinations or off‑target outputs. Overlong prompts may introduce noise.
Category: Mistakes
Why use chain-of-thought in Prompting?
Chain‑of‑thought encourages stepwise reasoning that improves complex problem solving. It must be applied responsibly to avoid leaking sensitive reasoning content.
Category: Relevancy
How do you evaluate prompt quality?
Use test suites, rubrics, and automatic metrics; measure correctness, consistency, and adherence to format. Iterate with A/B tests and error analysis.
Category: Requirements
What is the role of system prompts in Prompting?
System prompts set global instructions and tone, constraining model behavior across interactions. They complement user prompts and tool responses.
Category: Example
How does RAG complement Prompting?
RAG injects fresh, grounded context while prompting structures the request. Together they reduce hallucinations and improve factuality.
Category: Opinion
How should Prompting handle safety requirements?
Include explicit safety rules, disallowed behaviors, and references to policies. Test prompts for edge cases and adversarial inputs.
Category: Risk
Where to start learning Prompting?
Begin with vendor best practices, experiment with templates, and build a prompt library. Analyze failures systematically to refine.
Category: Requirements